ทวีตที่ปักหมุด
I wrote about the generative video editor I've been working on for the past year, and the progress so far: makeartwithpython.com/blog/a-year-of…
English
Kirk Kaiser
3.2K posts
Kirk Kaiser
@burningion
Python developer, author of Make Art with Python. On a mission to bring art and play to software development. Previously lead DevRel at Datadog.
Naples, FL เข้าร่วม Mart 2009
2.9K กำลังติดตาม1.4K ผู้ติดตาม
So there's two things: the broken state of OpenTelemetry auto instrumentation in the (I'm assuming) JS ecosystem (which may be valid, but also an opportunity as a vendor), and the net value proposition of Tracing.
Tracing done right should be a way to organizationally go an abstraction layer above the behavior of your specific code / system, and provide full context for a given workload.
Proper instrumentation should allow everyone to think in terms of all running services and their live behaviors / interactions for a given workload rather than a pile of information to be sorted through and filtered like logs, which usually require system specific knowledge to be maximally useful.
English
I couldn't really figure out how to articulate the message I wanted here, but I'm hoping this makes some of you reconsider if you need tracing (spans).
The worst software is complicated. The more we can simplify problems, the better it is for users (velocity, reliability), the better it is for developers (maintainability). Tracing is a complexity problem.
Recently that has led me to opt to use logs instead of traces, and because Sentry has trace connectivity on every data payload, it drastically simplifies my approach to production telemetry.
cra.mr/tracing-sucks
English
Wonder what percentage of market price discovery is being done by LLMs at this point and whether their sycophancy might be exploitable
English
@trishankkarthik He comes home to men stealing his wealth, harassing his wife, and actively trying to kill his kid. He arrives in disguise to assess the situation before revealing himself. Odysseus is a thinking and strategizing hero first.
English
@M5Stack How do I get notified when this is ready? And does it come with a battery? What sort?
English
Sometimes when a startup wants to escape from a bad organization that has some claim on their IP, it's worth rewriting their software from scratch in a "clean room." AI-assisted programming will make this much easier.
English
I am thrilled to share that we raised an oversubscribed $82M Fund III to lead pre-seed and seed rounds in Vertical AI, Deep Tech, Healthcare and Biotech startups in NYC and Boston.
To be exact, we raised $82,048,000 - a number that felt on-brand for a nerdy firm like ours that loves riddles, the future, Morse code, and Easter eggs.
We are immensely grateful to our LPs for their ongoing belief in our strategy and approach. We are equally grateful to the founders who have chosen to partner with us.
Here is 2048 Ventures Ventures Fund III in a nutshell:
- We back exceptional, ultra-competitive, visionary founders.
- We lead pre-seed and seed rounds with $500K - $3M checks.
- We invest in Vertical AI, Deep Tech, Healthcare and Biotech.
- We invest primarily in NYC and Boston.
- We seek businesses with strong data moats.
Read more about where we’re excited to invest:
2048.vc/blog/2048-vent…
If you are a builder and visionary looking for the highest conviction first-check investor, please reach out or re-share with your network!
English
@antirez Man there used to be this really cool place called the market where you could make a bet on the gap between what companies promised and what reality decided.
English
I don't understand how META AI ambitions are still trusted by markets. It is the company with the worst ever AI spending compared to the results they got. Partially because of the LLM skepticism inside, and partially because of wrong picking of the heads of the AI divisions.
English
@Afinetheorem @InsiderTakes The problem is that this line of thinking assumes the future will look like the past. This year has been a terribly volatile business environment in the United States in ways that feel like they’re accelerating.
English
@burningion @InsiderTakes The numbers are *worse* for Canada incorporated if you look at the tail. Source: I've been involved with Canada's biggest accelerator since it started, that saw half of Canada's unicorns in the past decade come through.
English
Garry from YC on why they require Singapore, Cayman, or US incorporation (physical HQ can be anywhere). A substantial literature now that both legal domicile and physical location matter for startup performance. The response to this should be "what policy do we have wrong"!
Garry Tan@garrytan
In YC’s 20 year history Canadian startups that reincorporated in the US have 2x the avg valuation of those that didn’t. And the ones at Unicorn or near it all reincorporated in Delaware I love Canada (born in Winnipeg). But don't let that get in the way of making a huge startup.
English
@InsiderTakes @Afinetheorem It only takes missing one Shopify to disprove this
English
@burningion @Afinetheorem Amazing that you thought of that and Garry Tan didn’t. He must be wrong about the returns to YC startups based on domicile.
English
@Afinetheorem Gary’s explained heuristics are unbelievably bad. Averages don’t matter when all the returns are at the tail.
English
@ekzhang1 That being said, we should still try, especially when it’s difficult.
English
@ekzhang1 I think there’s an underlying lack of consent in training data that makes building new systems from it in a good, human first way challenging. It’s tough to be default naively optimistic anymore about downstream implications of new tech because of that foundation.
English
I feel like there’s more cool software (useful, creative, well-designed, beautiful, etc) coming out every day than I ever remember. It’s strange that I also feel a bit… tired, or confused at times?
A few years ago I feel the industry didn’t really care about experience if you’re great at programming. An engineer can pretty much figure anything out in a new field within a few weeks-months, and by that time you’re an expert—that’s good enough. General coding knowledge is enough. Most software, companies (think Figma) took years of being small to build before they could scale; a bit of learning was part of the plan.
But I think, with everything going so fast now, timelines are shifting from years to months to weeks. There’s a pressure to be productive earlier, and even those few weeks of on-ramp on any given thing hurt execution speed. I hear more conversations about “have they done it before” rather than “how can we do something amazing?” — the predictability is sought after.
Meanwhile, software is getting written so quickly, sometimes I’d just like to take a step back from all of this and return a year later.
What we do should be in service to people; computers are how we learn, work together, create and collaborate, and software is how we create interactive media for people to discover. I’d like to continue being hopeful and drawing inspiration from these foundations. We’re really lucky, getting to create these digital artifacts that strike new wonder every day! No matter how industry or tools move.
(alt title: “eric gives himself a pep talk” lol)
English
Kirk Kaiser รีทวีตแล้ว
n config options can often be replaced by k options (with k << n) where k just lets the user express what thing they want to optimize for (typically some tradeoff between latency, throughput, and cost).
This is often very hard! But a 10x better user experience.
English
@venturetwins Some earlier ideas that might be interesting here: makeartwithpython.com/blog/a-year-of…
English
The absolutely worst place and situation I’ve ever been to. 😭😭😭
English
What are the best books you’ve read on navigating bureaucracy?
English
@zeeg Ah I think I see your point and this is probably more of a javascript artifact more than anything. Im doubtful of an LLM reimplementing numpy or pytorch on every inference.
English
@burningion I’m really only talking about the trivial ones - but there’s a lot of those. They’re also often not perfect fits for your project and contain more than you need. It’s already happening..
English
Tailwind isn’t open core
Tbh this whole thread is a fundamental misunderstanding of tailwinds situation, what open source is, and a good number of other things
Closed source won’t save you - code generation likely will replace much of open source in the coming years. There will be no more random small easily reproducible dependencies, as code generation will make it safer and more secure than relying on unmaintained code.
You can hate AI and still celebrate the win a reduction of the dependency graph will bring. No more left-pad.
Marc@MarcJSchmidt
All my new code will be closed-source from now on. I've contributed millions of lines of carefully written OSS code over the past decade, spent thousands of hours helping other people. If you want to use my libraries (1M+ downloads/month) in the future, you have to pay. I made good money funneling people through my OSS and being recognized as expert in several fields. This was entirely based on HUMANS knowing and seeing me by USING and INTERACTING with my code. No humans will ever read my docs again when coding agents do it in seconds. Nobody will even know it's me who built it. Look at Tailwind: 75 million downloads/month, more popular than ever, revenue down 80%, docs traffic down 40%, 75% of engineering team laid off. Someone submitted a PR to add LLM-optimized docs and Wathan had to decline - optimizing for agents accelerates his business's death. He's being asked to build the infrastructure for his own obsolescence. Two of the most common OSS business models: - Open Core: Give away the library, sell premium once you reach critical mass (Tailwind UI, Prisma Accelerate, Supabase Cloud...) - Expertise Moat: Be THE expert in your library - consulting gigs, speaking, higher salary Tailwind just proved the first one is dying. Agents bypass the documentation funnel. They don't see your premium tier. Every project relying on docs-to-premium conversion will face the same pressure: Prisma, Drizzle, MikroORM, Strapi, and many more. The core insight: OSS monetization was always about attention. Human eyeballs on your docs, brand, expertise. That attention has literally moved into attention layers. Your docs trained the models that now make visiting you unnecessary. Human attention paid. Artificial attention doesn't. Some OSS will keep going - wealthy devs doing it for fun or education. That's not a system, that's charity. Most popular OSS runs on economic incentives. Destroy them, they stop playing. Why go closed-source? When the monetization funnel is broken, you move payment to the only point that still exists: access. OSS gave away access hoping to monetize attention downstream. Agents broke downstream. Closed-source gates access directly. The final irony: OSS trained the models now killing it. We built our own replacement. My prediction: a new marketplace emerges, built for agents. Want your agent to use Tailwind? Prisma? Pay per access. Libraries become APIs with meters. The old model: free code -> human attention -> monetization. The new model: pay at the gate or your agent doesn't get in.
English